7. Noam Chomsky¹

7.1 Both inside and outside the discipline, Chomsky's work has fundamentally affected views of what of linguistics is or should be, and reopened issues many linguists had long thought were settled. On the jacket of Aspects of the Theory of Syntax (hereafter AT), a reviewer calls the ‘approach’ ‘truly fresh and revolutionary’; and Chomsky often stresses how it is different from, and better than, various alternatives. Yet many of his ideas are conservative in that they derive from traditional philosophy, grammar, and logic. The most ‘revolutionary’ aspect lies in his claims about how these ideas apply to language and linguistics (cf. 7.78, 95).

7.2 A skilful public debater, Chomsky intensifies the forensic and polemical aspects of the discipline by using theoretical arguments about ‘the nature of language’ to fortify his positions against competitors (cf. 9.3; 12²). He foregrounds points of contention even where he implicity agrees with or borrows from his adversaries, and uses a highly confident rhetoric for his ‘tentative’ views and proposals (cf. 7.85, 94).² His argumentation oscillates from intuitive reasoning and philosophical speculations on ‘the mind’, over to technical points drawn from formal language theory and from such sciences as biology and neurology. Due in part to this diversity of sources, his terminology and notation take on a strategic plurality of meanings (7.15, 28, 78, 83ff).

7.3 Chomsky turns away from ‘modern linguistics’ (cf. 7.5, 19, 30, 62, 75) and cites far earlier sources: Panini, Plato, and both rationalist and romantic philosophers, such as Rene Descartes (1647), Claude Faure Vaugelas (1647), Cesar Chesneau DuMarsais (1729), Denis Diderot (1751), James Beattie (1788), and Wilhelm von Humboldt (1836) (AT v, 4f, 49, 198f, 233f).³ He sees ‘the conception of linguistic structure that marked the origins of modern syntactic theory’ in the Grammaire generale et raisonnee (Lancelot & Arnaud 1660), which claimed that ‘aside from figurative speech’ the ‘“natural order of thoughts”‘ is ‘mirrored by the order of words'⁴ -- an idea Chomsky rejects as a ‘naive view of language structure’ blocking ‘a precise statement of regular processes of sentence formation’, though his notion of ‘deep structure’ being ‘interpreted’ via ‘semantic universals’ has a similar cast (AT 6, 117f, 137) (cf. 7.20, 71f; 11.84). Such allegiances brought ‘Cartesian linguisitics’ into fashion (cf. Chomsky 1966) in a discipline that had been rather hostile to philosophy (cf. 4.4ff, 19, 38, 51, 72; 6.3, 5, 13; 8.5, 17, 60, 8¹⁴; 9.3ff, 31; 13.16).

7.4 Another chief source is ‘traditional grammars’, staunchly defended for having ‘exhibited’ ‘linguistic processes’, ‘however informally’, and ‘given a wealth of information concerning the structural descriptions of sentences’ (AT 5). For Chomsky, this ‘information is ‘without question substantially correct, and is essential to any account of how the language is used or acquired’ (cf. 2.6; 6.49; 12.41, 88; 13.7); that such ‘grammars’ ‘have been “long condemned by professional linguists”‘ is deemed ‘irrelevant’ (AT 63f, 194) (cf. Dixon 1963). Admittedly, these ‘grammars were deficient in that they left unexpressed many of the basic regularities of language’, and never went ‘beyond the classification of particular examples to the stage of formulation of generative rules on any significant scale’; but this ‘defect’ is also found in ‘structuralist grammars’ (AT 5). Chomsky's grammatical terms in both SS and AT are not attributed to any source; language philosophers are cited only in theoretical arguments.

7.5 A source Chomsky roundly repudiates is ‘structural linguistics’, his chief competitor. His ‘discussion’ of workin this discipline leads the agenda of Syntactic Structures (hereafter SS) and is cited in AT as ‘unanswerable’ or at least ‘for the moment, not challenged’, whereupon he presumes that ‘the inadequacy’ of ‘structuralist grammars’ ‘for natural languages’ ‘has been established beyond any reasonable doubt’ (AT 54, 67). He simply ignores the merits of such grammars for describing a host of previously unrecorded languages (5.2; 13.56). He sees only their theoretical flaws and ‘no compensating advantage for the modern descriptivist reanalysis’, and even vows that ‘knowledge of grammatical structure cannot arise’ from the ‘operations’ used in that kind of ‘linguistics’ (AT 174, 57) (cf. 7.7, 24, 29, 33, 76, 96). Also, descriptivist fieldwork conflicts with his own plan to focus on English and to discount the formal diversity of languages as a ‘surface’ issue (cf. 7.18f, 41, 55, 62; 13.71).

7.6 Chomsky charges that ‘descriptivists’ rely unduly on induction: the ‘limitation-in-principle to classification and organization of data, to “extracting patterns” from a corpus of observed speech’ (AT 15) (cf. 4.7; 13.45). ‘Structural linguistics’ makes an ‘extremely strong demand’ on ‘discovery’ by ‘assuming that the techniques for discovering the correct hypotheses (grammar) must be based on procedures of segmentation and classification of items in a corpus’ (AT 202f) (cf. 13.30). Thus, ‘general linguistic theory’ seems to ‘consist only of a body of procedures’ which ‘determine’ the ‘restrictions on possible grammars’ and otherwise leave ‘the form of language unspecified’ (AT 52). ‘Proposals’ ‘attempt to state methods of analysis that an investigator might actually use’ ‘to construct a grammar of language directly from the raw data’ (e.g. Wells 1947; Bloch 1948:, Harris 1951, 1955; Hockett 1952; Chomsky 1953) (SS 52). ‘This goal’ can ‘lead into a maze’ of ‘complex analytic procedures’ unsuited to answer ‘many important questions about the nature of linguistic structure’ (SS 53).

7.7 In SS, Chomsky does ‘not deny the usefulness’ of ‘discovery procedures’ in ‘providing valuable hints to the practising linguist’, but only admonishes that ‘linguistic theory’ is not ‘a manual’ of ‘procedures’ (SS 55n, 59, 106) (cf. 6.38, 61; 8³; 12.33). In AT, however, he pointedly devalues the discovery of data, saying that ‘no adequate formalizable techniques are known for obtaining reliable information concerning the facts of linguistic structure’; and that ‘there is no reason to expect that reliable operational criteria for the deeper and more important notions of linguistic theory’ ‘will ever be forthcoming’ (AT 19). ‘Knowledge of the language, like most facts of interest and importance, is neither presented for direct observation nor extractable from data by inductive procedures of any known sort’ (AT 18). ‘Allusions’ to ‘“procedures of elicitation” or “objective methods” simply obscure the actual situation in which linguistic work’ must ‘proceed’ (AT 19) (cf. 5.10). So ‘theoretical’ ‘investigation of the knowledge of the native speaker can proceed perfectly well’ without ‘operational procedures’. ‘The critical problem is not a paucity of evidence but rather the inadequacy of present theories of language to account for masses of evidence’ ‘hardly open to serious question’ (AT 19f). These arguments lead to a saving of labour: ‘once we have disclaimed practical discovery procedures’, ‘certain problems’ likely to stir up ‘intense methodological controversy simply do not arise’ (SS 56). Indeed.

7.8 ‘The relation between the general theory and the particular grammars’ is where Chomsky's ‘approach’ ‘diverges sharply from many theories’ (SS 50). In his view, ‘a discovery procedure’ whereby ‘the theory’ ‘provides a practical and mechanical method for actually constructing the grammar, given a corpus of utterances’, is an unduly ‘strong requirement’ (SS 50f). ‘A weaker requirement’ is ‘a decision procedure’ whereby the ‘method’ ‘determines whether or not a grammar is’ ‘the best grammar’ for the ‘corpus’, without asking ‘how this grammar was constructed’. ‘Even weaker’ is ‘an evaluation procedure’ for telling which of ‘two proposed grammars’ (or ‘small sets of grammars’) is ‘better’. He thinks ‘it is unreasonable to demand of linguistic theory’ ‘more than a practical evaluation procedure’ (SS 52f). Having one would ‘guarantee significance’ for linguistics as one of the ‘few areas of science’ seeking a ‘practical mechanical method for choosing among several theories, each compatible’ with ‘the data’.

7.9 So a ‘theory may not tell us in any practical way’ how to ‘construct the grammar, but it must tell us how to evaluate’ and ‘choose’ (SS 54). Chomsky again saves labour by ‘never considering the question of how one might have arrived at the grammar’, and declaring ‘such questions’ ‘not relevant to the programme of research’ (SS 56) (cf. 13.38). ‘One may arrive at a grammar by intuition, guesswork’, ‘partial methodological hints’, ‘past experience’, etc.’ We might ‘give an organized account of useful procedures of analysis’, but these cannot be ‘formulated rigorously, exhaustively, and simply enough’ to ensure valid ‘discovery’. Instead, his ‘ultimate aim is to provide an objective, non-intuitive way to evaluate a grammar once presented’.

7.10 As Chomsky sets the task, ‘the grammarian must construct a description, and, where possible, an explanation, for the enormous mass of unquestionable data concerning the linguistic intuition of the native speaker, often himself’ (AT 20, i.s.) (cf. 6.15). Against much of earlier American linguistics, Chomsky vows that ‘linguistic theory is mentalistic’, ‘concerned with discovering a mental reality underlying actual behaviour’ (AT 4) (cf. 4.8; 13.57). He decries ‘behaviourism’ for consulting only ‘data’ and ‘expressing a lack of interest in theory and explanation’, due to ‘certain ideas’ (like ‘operationalism’ and ‘verificationism’) ‘in positivist philosophy of science’ (AT 193f). Instead, ‘linguistic theory’ should ‘contribute to the study of human mental processes and intellectual capacity’ (AT 46) (cf. 3.10ff; 5.69; 6.6; 7.10; 12.17ff, 22; 13.22). ‘Observed use of language or hypothesized dispositions to respond, habits, and so on, may provide evidence’ about the ‘mental reality, but surely cannot constitute the subject-matter of linguistics’ as ‘a serious discipline’ (AT 4).

7.11 ‘Introspective judgments of the informant’ cannot be ‘disregarded’ ‘on grounds of methodological purity’ without ‘condemning the study of language to utter sterility’ (AT 194). In SS, however, Chomsky had warned against ‘asking the informant to do the linguists’ work’ by eliciting a ‘judgment about his behaviour’; such ‘opinions may be based on all sorts of irrelevant factors’ (SS 97n) (cf. 13.49). SS also announced that ‘the major goal of grammatical theory is to replace’ the ‘obscure reliance on intuition by some rigorous and objective approach’ (SS 94; cf. SS 5, 56).⁵ But in AT, the prospect that ‘giving such priority to introspective evidence’ and ‘intuition’ might ‘exclude’ ‘linguistics’ from ‘science’ is downplayed as a ‘terminological question’ with ‘no bearing at all on any serious issue’ (AT 20). Though ‘the successful sciences’ are ‘concerned’ with ‘objectivity’, the latter ideal can, as ‘the social and behavioural sciences’ show, ‘be pursued with little gain in insight and understanding’. Chomsky's preferred model is ‘the natural sciences’, which ‘have sought objectivity as a tool for gaining insight, for providing phenomena’ to ‘suggest or test deeper explanatory hypotheses’, rather than ‘as goal in itself’ (cf. 2.13; 4.8, 18; 7.16; 9.112; 12.14, 49, 99; 13.11, 18). Regarding whether ‘a wider range and more exact description of phenomena is relevant’, Chomsky asserts that ‘in linguistics’, ‘sharpening the data by objective test’ is ‘of small importance’ for ‘new and deeper understanding of linguistic structure’ (AT 20f). ‘Many questions’ ‘today’ ‘do not demand evidence’ ‘unattainable without significant improvements in objectivity of experimental technique’ (i.a.).

7.12 Chomsky makes the declaration, soon to be famous, that ‘linguistic theory is concerned primarily with an ideal speaker-hearer in a completely homogeneous speech-community, who knows its language perfectly, and is unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic) in applying his knowledge of the language in actual performance’ (AT 3) (cf. 7.96; 9.5; 11.69; 12.44; 13.14, 18, 36, 13²). Chomsky thus draws an influential ‘distinction between competence, the speaker-hearer's knowledge of his language, and performance, the actual use of language in concrete situations’ (AT 4) (cf. 9.6; 13.39). ‘Only under the idealization’ of the ‘speaker-hearer’ ‘is performance a direct reflection of competence’, but not ‘in actual fact’. ‘A record of natural speech will show numerous false starts, deviations from rules, changes of plan in mid-course, and so on’. Chomsky's ‘distinction’ ‘is related to the langue-parole distinction of Saussure’ (2.20), but ‘rejects his concept of language as merely a systematic inventory of items, and returns to the Humboldtian conception of underlying competence as a system of generative processes’.

7.13 Chomsky subscribes to the ‘view’ ‘that the investigation of performance will proceed only so far as the understanding of underlying competence permits’ (AT 10). ‘The only concrete results’ and ‘clear suggestions’ so far for ‘a theory of performance’ have come from ‘models that incorporate generative grammars’ by making ‘assumptions about underlying competence’; and he finds it ‘difficult to imagine any other basis’ (AT 10, 15). Only within such models will ‘actual data of linguistic performance’ provide ‘evidence for determining the correctness of hypotheses about underlying structure, along with introspective reports by the native speaker or the linguist who has learned the language’ -- this being ‘the position universally adopted in practice’ (AT 18). He cites ‘observations concerning limitations on performance imposed by organization’ and ‘bounds of memory’, or ‘concerning exploitations of grammatical devices to form deviant sentences’ (AT 10) (cf. 7.41).

7.14 Chomsky stipulates that ‘a theory of linguistic intuition’, a ‘grammatical description’, or ‘an operational procedure’ ‘must be tested for adequacy’ ‘by measuring it against’ ‘the tacit knowledge’ it tries to ‘describe’ (AT 19ff) (cf. 7.24, 29, 91). But even if it is ‘the ultimate standard’ of ‘accuracy’ and ‘significance’ for both ‘grammars’ and ‘tests’, this ‘tacit knowledge may very well not be immediately available to the user of the language’ (13.49). Precisely due to the ‘elusiveness of the speaker's tacit knowledge’, the linguist is needed to ‘guide and draw it out’ in ‘subtle ways’ (AT 24). By ‘adducing’ examples, we can ‘arrange matters’ so that people's ‘linguistic intuition, previously obscured, becomes evident to them’. Chomsky's favoured tactic is to show that an isolated ‘sentence’ (like ‘“flying planes can be dangerous”’) can be ‘assigned’ more than one ‘interpretation’ ‘by the grammar’, although ‘in an appropriate’ ‘context, the listener will interpret it immediately in a unique way’ and ‘may reject the second interpretation’, if ‘pointed out to him, as forced or unnatural’ (AT 21) (cf. 7.53, 61, 79, 82; 13.39).

7.15 Along similar lines, Chomsky finds it ‘obvious’ that ‘every speaker of a language has mastered and internalized a generative grammar that expresses his knowledge of his language’ (AT 8). Indeed, Chomsky ‘uses the term “grammar” with a systematic ambiguity’ both for ‘the native speaker's internally represented “theory of his language”‘ and for ‘the linguist's account of this’ (AT 25) (cf. 7.28, 78; 13.45). Yet people may not be ‘aware of the rules’, or even able to ‘become aware’, nor are their ‘statements about their intuitive knowledge’ ‘necessarily accurate’ (AT 8) (13.49). ‘Any interesting generative grammar will be dealing, for the most part, with mental processes’ ‘far beyond the level of actual or even potential consciousness’ (AT 8; cf. AT 59; 2¹⁶; 13.49). Since ‘the speaker's reports and viewpoints about his behaviour and competence may be in error’, a ‘grammar attempts to specify what the speaker knows, not what he may report’. Chomsky sees a ‘similarity’ here to ‘a theory of visual perception’ trying to ‘account for what persons actually see and the mechanisms that determine this rather than for their statements about what they see and why, though these statements may provide useful, in fact compelling evidence’ (AT 8f) (cf. 7.35, 89). However, seeing objects or scenes differs from ‘seeing’ language, if in ‘linguistics’ ‘the viewpoint’ ‘creates the object’, as Saussure asserted (2.9; 13.58).

7.16 Further analogies with other sciences are strategically drawn. Using a ‘grammar’ to ‘reconstruct formal relations among utterances in terms’ of ‘structure’ and to ‘generate all grammatically “possible” utterances’ is ‘analogous’ to using ‘chemical theory’ to ‘generate all physically possible compounds’ (SS 48) (cf. 2.82; 4.8; 5.28; 13.18). A ‘chemical theory’ ‘serves as a theoretical basis for techniques of qualitative analysis and synthesis of specific compounds, just as a grammar’ supports ‘the investigation’ of the ‘analysis and synthesis of particular utterances’. Or in physics, ‘any scientific theory is based on a finite number of observations’, which it ‘relates’ and ‘predicts’ ‘by constructing general laws in terms of such hypothetical constructs as’ ‘“mass and electron”‘ (SS 49) (4.8, 71; 5.66; 7.36; 6.62; 8.49; 13.43). ‘Similarly, a grammar of English is based on a finite corpus of utterances (observations)’, ‘contains grammatical rules (laws) stated in terms of phonemes, phrases, etc.’ ‘(hypothetical constructs)’ and ‘expresses structural relations among sentences of the corpus and the indefinite number of sentences generated by the grammar’ (‘predictions’). Such ‘physical’ analogies stand out in a mentalist approach, especially one whose main ‘metaphor’ seems to be mathematics (cf. 7.93; 13.10, 15, 18).

7.17 Chomsky proposes to ‘construct a formalized general theory of linguistic structure and to explore’ its ‘foundations’, hoping to ‘fix in advance for all grammars’ the way they are ‘related to a corpus of sentences’ (SS 5, 14). ‘We can attempt to formulate as precisely as possible both the general theory and the set of associated grammars’ (SS 50).⁶ His ‘motivation’ is not just a ‘mere concern for logical niceties or desire to purify well-established methods of linguistic analysis’ (SS 5). ‘Precisely constructed models’ ‘can play an important role, both negative and positive, in the process of discovery’ (cf. 7.98). ‘By pushing a precise but inadequate formulation to an unacceptable conclusion’ and ‘exposing the source’, we can ‘gain a deeper understanding of the linguistic data’. Or, ‘positively, a formalized theory may automatically provide solutions for problems’ for which it wasn't ‘originally designed’. To ‘determine the fundamental underlying properties of successful grammars’, ‘linguists’ should ‘recognize the productive potential’ of ‘rigorously stating a proposed theory and applying it strictly to linguistic material’ without ‘ad hoc adjustments or loose formulations’ (SS 11, 5). In any event, ‘neither the general theory nor the particular grammars are fixed for all time’ (SS 50). ‘Progress and revision may come’ from ‘new facts’ or ‘purely theoretical insights’, i.e., ‘new models for linguistic structure’.

7.18 In fine, Chomsky's priorities sound like Hjelmslev's:⁷ ‘a theory of language must state the principles interrelating theoretical terms’ and ‘ultimately must relate this system to potential empirical phenomena, to primary linguistic data’ (AT 208).⁸ We should ‘describe the form of grammars (equivalently, the nature of linguistic structure)’ and ‘the empirical consequences of adopting a certain model’ (SS 56). ‘The ultimate outcome’ ‘should be a theory of linguistic structure in which the descriptive devices utilized in particular grammars are presented and studied abstractly’, without ‘reference to particular languages’ (SS 11, 50) (cf. 6.11). Having ‘data’ from ‘relatively few languages’ ‘is not particularly disturbing’, because the ‘conditions’ of ‘a single language may provide significant support’ for ‘some formal property’ belonging to ‘general linguistic theory’ (AT 209). Chomsky sees ‘an important advance in the theory of language’ when an apparent ‘peculiarity of English’ is made ‘explicable’ by ‘a general and deep empirical assumption about the nature of language’, which can be ‘refuted if false’ by ‘grammars of other languages’ (AT 36). This step fits the hope of ‘replacing’ ‘assertions about particular languages’ with ones about ‘language in general’, so that ‘features of grammars in individual languages can be deduced’ (AT 46). Conversely, ‘the difficulty or impossibility of formulating certain conditions within’ a ‘theory of grammar’ may signal that they are ‘general’ for ‘the applicability of grammatical rules rather than aspects of the particular language’ (AT 209).

7.19 So against ‘modern linguistics’, Chomsky calls for ‘the grammar of a particular language’ ‘to be supplemented by a universal grammar’ ‘expressing deep-seated regularities’ (AT 6) (cf. 2.10; 3.67; 4.4, 71f, 74; 5.44; 6.5, 10, 34; 7.22, 29, 33f, 45, 55, 62, 65, 71, 78f, 91, 93, 7¹⁰, 7³², 7³⁹; 8.19, 60, 86; 9.3, 25, 60; 12.94; 13.18). Indeed, ‘the main task of linguistic theory must be to develop an account of linguistic universals’ (AT 28). He contests the ‘commonly held’ view ‘that modern linguistic and anthropological investigations have conclusively refuted the doctrines’ of ‘universal grammar’ (AT 118) (cf. 7.55, 62). ‘Linguistic theory’ may yet ‘develop an account of linguistic universals’ that reveals ‘the properties of any generative grammar’ and is ‘not falsified by the diversity of languages’ (AT 27f). An important (and labour-saving) corollary is that ‘universal’ ‘regularities’ ‘need not be stated in the grammar’ of a ‘language’ but ‘only in general linguistic theory as part of the definition of the notion “human language”‘ (AT 6, 117; cf. AT 35f, 112, 144, 168, 225; 7.65, 72).

7.20 ‘Universals’ fall into two kinds. The ‘substantive’ ones, a ‘traditional concern of general linguistic theory’, involve ‘the vocabulary for the description of language’ and state that ‘items of a particular kind in any language must be drawn from a fixed class’ (AT 28). The best case comes from phonology, the old standby of linguists, namely the ‘theory of distinctive features’, ‘each of which has a substantive acoustic-articulatory characterization independent of any particular language’ (cf. 2.70; 5.42f; 6.54; 7.71, 7¹⁰; 8³⁶; 13.26). Another case is the assumption of ‘traditional universal grammar’ that ‘fixed syntactic categories (Noun, Verb, etc.) can be found in the syntactic representations of the sentences of any language’ and ‘provide the general underlying syntactic structures of each language’. ‘More abstract’ and ‘recently’ studied are the ‘formal linguistic universals’ stating ‘formal conditions’ such as ‘the character of the rules’ in ‘grammar’ and the ways’ they are ‘interconnected’ (AT 29). Such ‘deep-seated formal universals’ ‘imply that all languages are cut to the same pattern’, though without any ‘point-by-point correspondence’ (AT 30).⁹ These ‘formal constraints’ ‘may severely limit the choice’ ‘of a descriptive grammar’. Not surprisingly, Chomsky's favoured candidate here is ‘that the syntactic component of a grammar must contain transformational rules’; and that even ‘the phonological component’ ‘consists of a sequence of rules, a subset of which’ is ‘a transformational cycle’ (AT 29). This move too saves labour: if ‘the transformational cycle is a universal feature of the phonological component, it is unnecessary, in grammar of English, to describe’ the ‘functioning of phonological rules that involve syntactic structure’ (AT 35).¹⁰ A new rationale for separating levels?

7.21 Therefore, ‘our problem is to develop and clarify the criteria for selecting the correct grammar for each language, that is, the correct theory of this language’ (SS 49). ‘A linguistic theory is descriptively adequate’ and ‘empirically significant’ if it ‘makes available for each language’ a ‘grammar’ (or ‘a class of grammars’), ‘correctly’ ‘states the facts’, ‘describes the intrinsic competence of the idealized native speaker’ for ‘understanding arbitrary sentences’, and ‘accounts for the basis of that achievement’ (AT 24, 27, 34, 40, i.s.). ‘External conditions of adequacy’ hinge on whether ‘the sentences generated’ are ‘acceptable to the native speaker’, but apply to a ‘merely descriptive’ ‘theory’ (SS 49f, 13, 54). ‘Internal grounds’ constitute a ‘much deeper’ and more ‘principled’ ‘level’ of adequacy (SS 13; AT 41, 27). For the moment however, Chomsky simply ‘tests the adequacy’ of an ‘apparatus’ by ‘applying it directly to the description of English sentences’ (SS 34) (cf. 7.80). In fact, he makes do with the ‘weak test’ of ‘a certain number of clear cases’; a ‘strong test’ would require these to ‘be handled properly for each language by grammars all’ ‘constructed by same method’ (SS 14) (cf. 7.42; 13.40).

7.22 ‘The structural descriptions assigned to sentences by the grammar’ and its ‘distinctions between well-formed and deviant’ ‘must correspond to the intuition’ of the ‘speaker, whether or not he’ is ‘aware’ of it (AT 24) (cf. 7.11; 13.50). Yet a ‘descriptively adequate’ theory may ‘provide such a wide range of potential grammars’ that ‘no formal property distinguishes’ them, or may ‘leave unexpressed’ ‘the defining properties of natural languages’ that ‘distinguish’ them ‘from arbitrary symbol systems’ (AT 35f). So ‘the major endeavour of the linguist must be’ to ‘enrich the theory of linguistic form’ and ‘restrict the range of possible hypotheses by adding structure’, ‘constraints, and conditions’ to ‘the notion “generative grammar”‘ and ‘to reduce the class of attainable grammars’ until ‘a formal evaluation measure’ can be applied (AT 35, 46, 41). ‘This requires a precise and narrow delimitation of the notion “generative grammar” -- a restrictive and rich hypothesis’ about ‘universal properties that determine the form of language’ (AT 35). ‘Given a variety of descriptively adequate grammars’ we need to discover if they are ‘unique’ or share ‘deep underlying similarities attributable to the form of language as such’. Only the latter discovery yields ‘real progress in linguistics’.

7.23 Hence, Chomsky deems it ‘crucial for the development of linguistic theory’ to ‘pursue’ ‘much higher goals’ than descriptive adequacy’, even ‘utopian’ ones (AT 24f). He envisions ‘explanatory adequacy’ when ‘a linguistic theory succeeds in selecting a descriptively adequate grammar on the basis of primary linguistic data’ and in relating ‘an explanation of the intuition of the native speaker’ to ‘an empirical hypothesis about the innate predisposition of the child’ (AT 25ff; cf 7.27f). Instead of ‘gross coverage of a large mass of data’, which is not ‘an achievement of any theoretical interest or importance’ (7.6f, 10f), ‘linguistics’ should ‘discover a complex of data that differentiates between conflicting conceptions of linguistic structure’ by showing which ones ‘can explain’ the ‘data’ via ‘some empirical assumption about the form of language’ (AT 26). Even for ‘descriptive adequacy’, an ‘explanatory theory of the form of grammar’ provides a ‘main tool’, because ‘the choice’ is ‘always underdetermined by data’ and because ‘relevant data’ from ‘successful grammars for other languages’ can be collated (AT 41) (13.43). Though both ‘unrealized goals’, ‘descriptive’ and ‘explanatory adequacy’ are ‘crucial at every stage of understanding linguistic structure’ (AT 36, 46).

7.24 Since ‘all concrete attempts to formulate an empirically adequate linguistic theory’ leave ‘room for mutually inconsistent grammars’, an ‘evaluation measure’ might be drawn from ‘language acquisition’ (AT 37). For Chomsky, ‘a theory of language’ can in fact ‘be regarded as a hypothesis’ about the ‘“language-forming capacity” of humans’ and ‘language learning’ (AT 30). We can thus ‘formulate’ ‘problems of linguistic theory’ and ‘language learning’ ‘as questions about the construction of a hypothetical language acquisition device’ or ‘model’ (AT 47, 25). The ‘child’ or ‘device’ ‘constructs a theory’ that ‘specifies its tacit competence, its knowledge of the language’, by ‘devising a hypothesis compatible with presented data’ of ‘performance’ (AT 32 36, 4). This feat sounds like a task for a linguist, and Chomsky makes the parallel explicit (cf. 7.88f). On one side, ‘the theorist’ is ‘given an empirical pairing of collections of primary linguistic data with certain grammars’ ‘in fact constructed by people’; on the other side, the child confronts ‘primary linguistic data consisting of signals, classified as sentences and non-sentences’, that are ‘paired with structural descriptions’ (AT 38, 47, 32). For this purpose, ‘the child has developed and internally represented a generative grammar’: ‘a theory’ and ‘a system of rules that determine how sentences are to be formed, used, and understood’ (AT 25).

7.25 ‘In part, such data determine’ which ‘possible language’ ‘the learner is being exposed’ to (AT 33). But the child's ‘internalized grammar goes far beyond the presented primary linguistic data’, ‘and is in no sense an “inductive generalization”‘ (AT 33, 25) (cf. 7.5ff, 30, 34, 93). Because apart from occasional ‘corrections of learners’ attempts’ by ‘the linguistic community’, ‘no special care is taken to teach’ ‘children’, the latter ‘must have the ability to “invent” a generative grammar that defines well-formedness and assigns interpretations to sentences even though linguistic data’ are ‘deficient’ (AT 31, 201). Indeed, Chomsky's idealizations imply that data in ‘actual speech’ are not merely ‘finite’, ‘scattered’, and ‘restricted in scope’, but also ‘degenerate in quality’, replete with ‘non-sentences’, ‘fragments, and deviant expressions’ (AT 43f, 31, 58, 201, 25; cf. 7.12; cf. 9.4f). He compares this factor with the ‘traditional view’ that ‘“the pains”‘ of ‘“conversation”‘ lie in ‘“extricating”‘ ‘“the thought from the signs or words which often agree not with it” (Cordemoy 1667)’ (AT 201), but his own concept of well-formedness is unrelated to the deviousness or overelaboration Cordemoy probably had in mind.

7.26 The child again resembles a Chomskyan linguist if ‘as a precondition for language learning, he must possess a linguistic theory that specifies the form of the grammar of a possible human language’ plus ‘a strategy for selecting a grammar’ ‘compatible’ with the ‘data’ (AT 25). Chomsky envisions the ‘child’ applying ‘a class of possible hypotheses about language structure’ and ‘determining what each’ ‘implies for each sentence’ so as to ‘select one of the presumably infinitely many hypotheses’ ‘compatible with the given data’ (AT 30, 45). Perhaps ‘an acquisition model’ entails a ‘strategy for finding hypotheses’ by ‘considering ‘only grammars’ above ‘a certain value’ before ‘language learning can take place (AT 203).¹¹ Or, ‘it is logically possible that the data’ are so ‘rich’ and ‘the class of potential grammars’ so ‘limited’ that only a ‘single permitted grammar will be compatible with the available data’ (AT 28). Or, even ‘a system not learnable by a language acquisition device’, which ‘is only one component of the total system of intellectual structures’ applied to ‘problem-solving and concept formation’, might be ‘mastered by a human some other way’; but Chomsky ‘expects ‘a qualitative difference’ in how ‘an organism’ ‘approaches’ ‘languagelike’ ‘systems’ (AT 56).

7.27 As ‘the most interesting and important reason for studying descriptively adequate grammars’ and for ‘formulating and justifying a general linguistic theory’, Chomsky states this ‘issue’: that ‘the general features of language structure reflect not so much the course of one's experience’ as the ‘capacity to acquire knowledge’ by ‘innate ideas’ and ‘principles’ (AT 59, 27, 78). Invoking ‘the traditional belief that “the principles of grammar form an important” “part of the philosophy of the human mind”‘ (Beattie 1788) (AT 59), he postulates a ‘structure’ that ‘pertains to the form of language in general’ and ‘reflects what the mind brings to the task of language acquisition, not what it discovers or invents’ during ‘the task’ (AT 59, 117). He hopes ‘linguistic theory’ can thereby ‘contribute to the study of human mental processes and intellectual capacity’ via ‘the abilities that make language learning possible under empirically given limitations of time and data’ (AT 46f, 31) (cf. 7.10; 13.22).

7.28 So ‘a long-range task for general linguistics’ is an ‘account of this innate linguistic theory’ forming ‘the basis for language learning’, although Chomsky short-circuits this task with another ‘systematic ambiguity’: using ‘“theory of language”‘ for ‘both the child's innate predisposition’ and for ‘the linguist's account’ (AT 25) (cf. 7.15, 78). A ‘hypothesis concerning the innate predisposition of the child’ can help ‘explain the intuition of the native speaker’ and can be ‘falsified’ when ‘it fails to provide’ a ‘grammar’ for ‘data from some other language’ (AT 25f). An ‘extremely strong’ ‘claim’ is involved here ‘about the innate capacity’ or ‘predisposition’ ‘of the child’ (AT 32) (cf. 7.93). It ‘is maintained’ ‘that the child has an innate theory of potential structural descriptions’ ‘rich and fully developed’ enough to ‘determine from a real situation’ ‘which structural descriptions may appropriate to a signal’ that ‘occurs’; and that this can be done ‘in advance of any assumption as to the linguistic structure of this signal’ (AT 32f) (cf. 6⁶).

7.29 ‘A theory of language’ along these lines ‘should concern itself’ with ‘linguistic universals’, a ‘theory’ or ‘tacit knowledge’ of which is ‘attributed’ ‘to the child’ (or to the ‘language acquisition device’) (AT 30, 27, 55f) (cf. 7.23, 89). ‘The child approaches the data with the presumption that they are drawn from a language’ of an ‘antecedently well-defined type’, and ‘determines which of the humanly possible languages is that of the community’ (AT 27). A ‘theory’ with ‘universals’ ‘implies that only certain kinds of symbolic systems can be acquired and used as languages’, whereas ‘others are beyond it’ (AT 55). We thus arrive at ‘the question: what are the initial assumptions’ about ‘the nature of language that the child brings’ and how ‘specific is the innate schema (the general definition of “grammar”) that gradually becomes more explicit and differentiated as the child learns the language?’ (AT 27).

7.30 ‘Acquisition of language’ offers ‘a special and informative case’ for ‘a more general’ ‘discussion’ of ‘two lines of approach to the acquisition of knowledge’ (AT 47). ‘Empiricism’ (spurred by ‘18th century struggles for scientific naturalism’) ‘assumes that the structure of the acquisition device is limited to certain elementary “peripheral processing mechanisms”‘, e.g. ‘an innate “quality space”, or to a set of primitive unconditioned reflexes’ or (for ‘language’) a set of ‘“aurally distinguishable components”‘; plus ‘elementary’ ‘analytical data-processing mechanisms or inductive principles’, e.g., ‘weak principles of generalization’ or (for language) ‘principles of segmentation and classification’ like those of ‘modern linguistics’ (AT 47, 51, 58f, 207) (cf. 7.5ff, 25, 33, 96; cf. 13.26, 30). ‘A preliminary analysis of experience is provided’ by those ‘mechanisms’ as ‘concepts and knowledge are acquired’ through ‘inductive principles’ (AT 48). Such claims are said to portray ‘language’ as ‘an adventitious construct’ (AT 51). Moreover, Chomsky complains that ‘empiricist views have generally been formulated in such an indefinite way’ as to be ‘next to impossible to interpret’, ‘analyse, or evaluate’ (AT 204). Skinner's (1957) account (reviewed in Chomsky 1959) is ‘grossly, obviously counter to fact’ if ‘terms like “stimulus”, “reinforcement”, “conditioning”, etc.’ are used ‘as in experimental psychology’; if they are just ‘metaphoric extensions’, we get only ‘a mentalist account differing from traditional ones’ by ‘the poverty of the terminology’ and not by any ‘scientific’ quality.¹²

7.31 ‘Rationalism’ is a ‘different approach’ to the ‘acquisition of knowledge’ and holds that ‘innate ideas and principles’, ‘fixed in advance as a disposition of mind’, ‘determine the form of acquired knowledge’ in a ‘restricted and highly organized way’ (AT 48, 51). A main inspiration is ‘seventeenth-century rationalist philosophy’ (AT 49). Descartes (1647) said that ‘“innate ideas”‘ about ‘“movements and figures”‘, ‘“pain, colour, sound”‘, etc. -- as well as ‘such notions as that things equal to same thing are equal to each other’ -- arise from ‘“our faculty of thinking”‘, not ‘“from external objects”‘ (AT 48). ‘The Port-Royal Logic (Arnauld [& Nicole] 1662)’ made ‘the same point’: ‘no idea’ ‘in our minds has taken its rise from sense’, because ‘ideas’ have ‘rarely any resemblance to what takes place in the sense and in the brain’, and ‘some have no connection with any bodily image’ (AT 49f). ‘Lord [Edward] Herbert [of Cherbury] (1624) maintained that’ without ‘innate ideas and principles’, ‘“we should have no experience”‘ or ‘“observations”‘, nor ever ‘“distinguish between things or grasp any general nature”‘ (AT 49). Leibniz (1873[1702-03]) declared ‘the senses ‘necessary’ but ‘not sufficient’ for ‘actual knowledge’, furnishing only ‘examples, i.e. particular’ ‘truths’, whereas ‘the truths of numbers’, i.e. ‘all arithmetic and geometry, are in us virtually’ to ‘set in order what we already have in the mind’ (AT 50). ‘Necessary truths must have principles whose proof does not depend on examples nor consequently’ on ‘the senses’, and which ‘form the soul’ of ‘our thoughts’, ‘as necessary thereto as the muscles’ for ‘walking’.

7.32 Descartes’ claim that ‘corporeal movements’ ‘reach our mind’ ‘from external objects’ and ‘cause’ it to ‘envisage these ideas’ (AT 48) matches Chomsky's stipulation that ‘innate mechanisms’ must be ‘activated’ by ‘appropriate stimulation’ and are then ‘available for interpretation of the data of sense’ (AT 48, 51). ‘The rationalist view’ implies that ‘one cannot really teach language, but can only present the conditions under which it will develop spontaneously in the mind’ (cf. Humboldt 1836) (but cf. 12.17, 11¹¹). ‘Thus the form of a language, the schema for its grammar, is to a large extent given’, though it must be ‘set into operation’ by ‘experience’ (AT 51).

7.33 ‘Empiricism’ is surprisingly accused of being not ‘“scientific”‘ but more ‘dogmatic and aprioristic’ than ‘rationalism’ in saying that ‘arbitrarily selected data-processing mechanisms’ ‘are the only ones available’ (AT 207). ‘Rationalism’, to which Chomsky finds it ‘difficult to see’ ‘an alternative’, is praised for holding ‘no preconceptions’ about ‘the internal structure’ of the ‘input-output device’, and only ‘studying uniformities in the output’, i.e. ‘universals’ (AT 48) (but cf. 7.93). If ‘rationalism’ does ‘not show how internal structure arises’, neither does ‘empiricism’ (AT 206). The rival is further outflanked by putting physics and biology, of all things, on the innatist side: we cannot ‘take seriously a position that attributes a complex human achievement to months or at most years of experience rather than to millions of years of evolution or to principles of neural organization’ ‘even more deeply grounded in physical law’ (AT 59).

7.34 Thus, even though ‘the empiricist notion’ is ‘the prevailing modern view’, Chomsky avers that ‘a hypothesis about initial structure’ able to ‘account for acquisition of language’ cannot fit ‘preconceptions’ from ‘centuries of empiricist doctrine’ -- ‘implausible to begin with’, ‘without factual support’, and ‘hardly consistent’ with how ‘animals or humans construct a ‘theory of the external world’ (AT 51, 58). He raises the prospect of ‘testing’ the two sets of ‘principles’ against ‘those we in fact discover’ in ‘real languages’, or gauging the ‘feasibility’ for ‘producing grammars within the given constraints of time and access’ and the ‘observed uniformity of output (AT 53f). But his own verdict is already decided: ‘empiricist theories about language acquisition are refutable wherever they are clear’, or else ‘empty’, just as ‘evidence’ about ‘language acquisition’ ‘shows clearly that taxonomic views’ are ‘inadequate’ because ‘knowledge of grammatical structure cannot arise’ by ‘inductive operations’ ‘developed in linguistics, psychology, or philosophy’ (AT 54, 57) (cf. 7.5ff, 24, 96; 13.45). In contrast, ‘the rationalist approach exemplified’ by ‘transformational grammar seems to have proven productive’ and ‘fully in accord with what is known about language’, and to offer the ‘hope of providing a hypothesis about the intrinsic structure of a language acquisition system’ (AT 54). Chomsky thereby preserves the hope that his ‘theory’ with its ‘linguistic universals’ may grow ‘rich and explicit enough to account for the rapidity and uniformity of language learning, and the remarkable complexity and range of generative grammars that are the product’, whereas ‘taxonomic linguistics’ can make no such ‘empirical claim’ (AT 28, 52f).

7.35 ‘The problem of mapping the intrinsic cognitive capacities of an organism and identifying the systems of belief and the organization of behaviour that it can readily attain should be central to experimental psychology’ (AT 56f). But due to ‘the atomistic and unstructured framework’ of ‘empiricist thinking’, ‘learning theory’ has instead ‘concentrated’ on the ‘marginal topic’ of ‘species-independent regularities in acquisition’ of ‘a “behavioural repertoire” under experimentally manipulable conditions’ (AT 205, 57). ‘Attention’ has focused on ‘tasks’ ‘extrinsic to cognitive structure’ ‘that must be approached in a devious, indirect, and piecemeal fashion’ (AT 57) (11.92). Chomsky is a bit inconsistent here. To discredit ‘empiricism’, he dismisses ‘comparisons with species other than man’ because ‘every species has highly specialized cognitive capacities’ and ‘language’ is ‘a human creation’ ‘reflecting intrinsic human capacity’ (AT 206, 59) (cf. 3.15; 4.28; 8.27; 12.10; 13.12, 18). The ‘analysis of stimuli’ ‘provided’ by ‘peripheral processing in the receptor systems or in lower cortical centres’ is ‘specific to the animal's life-space’ and ‘behaviour patterns’ (AT 205). Yet to discredit the ‘view that all knowledge is derived solely by the senses’ via ‘association and “generalization”‘, Chomsky denies that ‘man is’ ‘unique among animals’; and to make a point about ‘situational context’ (7.88), he turns to ‘animal learning’ (e.g., ‘depth perception in lambs’ ‘facilitated by mother-neonate contact’ upon which ‘the nature of the lamb's “theory of visual space”‘ does not ‘depend’) (AT 59, 34) (cf. 7.15).

7.36 All these arguments about mind, thought, and learning were absent from Chomsky's advocacy in SS. There, his ‘basic requirement’ for ‘any conception of linguistic structure’ was merely that ‘the grammar of English become’ ‘more simple and orderly’ and yet be able to ‘generate exactly the grammatical sentences’ (SS 68, 54n). The claim that his own approach can ‘simplify’ both ‘theory’ and ‘grammar’ has remained a major theme (SS 37, 41, 47, 55, 58, 65, 72, 106; AT 17, 87, 134, 136, 144, 202, 224) (cf. 7.39f, 42f, 47f, 50f, 54). He also stresses the ‘simplicity’ of very specific areas or tactics (usually ‘transformations’), and judges what is ‘simpler’ or the ‘simplest’ (SS 18, 84, 80; AT 62; SS 14, 56f, 74f, 85, 107, AT 55). Indeed, SS said Chomsky's ‘sole concern’ was ‘to decrease the complexity of the grammar’ ‘of English’ and make it ‘simpler than any proposed alternative’ (SS 83f, i.a.). Admittedly, ‘the notion of “simplicity”‘ ‘was left unanalysed’; Chomsky foresaw it being ‘defined within linguistic theory’, and he compared ‘choosing’ it to ‘determining the value of a physical constant’ as a ‘matter with empirical consequences’ (SS 103, AT 37f) (cf. 7.16). Instead of ‘simplifying one part’ and ‘complicating others’, ‘the right track’ lies where ‘simplifying one part leads to’ ‘simplifying others’ (SS 56). But SS proposed to ‘simplify the grammar’ ‘by formulating rules of a more complex type’ ‘than immediate constituent analysis’;¹³ and AT supports its arguments about ‘language learning’ by pointing to ‘the remarkable complexity’ of ‘generative grammars’ (SS 41; AT 28; cf. 7.50).

7.37 Simplicity was also a key point in Chomsky's original ‘contention that the conceptions of phrase structure are fundamentally inadequate and that the theory of linguistic structure must be elaborated’ through ‘transformational analysis’ (SS 69). Though not using ‘the strongest proof of the inadequacy of a linguistic theory’ (‘that it literally cannot apply to natural language’), Chomsky did apply the ‘weaker’ one that it ‘can apply only clumsily, in complex, ad hoc, unrevealing’ ways, and that ‘fundamental formal properties of natural language cannot be utilized to simplify grammars’ (SS 34). His plan was to compare ‘three models for linguistic structure’ and ‘their limitations’: a ‘communication theoretic model’, an ‘“immediate constituent”‘ or ‘phrase structure’ model, and a ‘transformational model’, and to show that the first two ‘leave gaps in linguistic theory’ and ‘cannot properly serve the purposes of grammatical description’, unless they are made ‘so hopelessly complex that they will be without interest’ (SS 6, 23, 41, 44, 18-49) (cf. 7.94).

7.38 The ‘communication theoretic model’, outlined by Shannon and Weaver (1949), implies only ‘a minimal linguistic theory’ (SS 18f, 34). ‘A machine’ ‘can be in any one of a finite number of different internal states’, and ‘switches’ among them ‘by producing a certain symbol’, e.g. ‘an English word’ (SS 18f). The ‘sequence’ from ‘an initial’ to ‘a final state’ is ‘a “sentence”‘ (i.r.). A ‘language produced in this way’ is ‘a finite state language’ and the ‘machine itself a finite-state grammar’ or a ‘“Markov process”‘ (SS 19ff). ‘To complete’ the ‘model’, ‘we assign a probability to each transition from state to state’, ‘calculate the uncertainty’, and thus ‘the “information content”‘ (cf. 7.90; 9²³). Plainly, ‘English is not a finite state language’. Its ‘symbols’ may not be ‘consecutive’, but ‘embedded’; and ‘Markov process models’ cannot ‘account for the ability of a speaker of English to produce and understand new utterances’ (SS 21ff). Imposing ‘arbitrary limitations’, e.g., that ‘sentences’ must be ‘less than a million words’, won't help, because in ‘English’ some ‘processes have no finite limit’.

7.39 A ‘description in terms of phrase structure’ is ‘more powerful’ than a ‘finite state’ one (SS 30f). Chomsky fits this second model to the ‘conception of linguistic structure’ of the last ‘half a century’, namely ‘the “taxonomic” view’ he wants to supplant (AT 88). He ‘does not know’ if ‘English itself is literally outside the range of analysis’ by ‘phrase structure’, but finds such grammars ‘certainly inadequate in the weaker sense’, e.g., in failing to ‘handle’ ‘discontinuous elements’ (SS 34, 38, 41, 75) (cf. 8.61; 13.28). Still, such ‘grammars’ are ‘quite adequate for a small part of the language’, while ‘the rest’ ‘can be derived by repeated application of a rather simple set of transformations to the strings given by the phrase structure grammar’ (SS 41fn) (cf. 7.62). In contrast, if we merely ‘extended phrase structure to cover the entire language directly, we would lose the simplicity of limited’ versions (SS 42).

7.40 We can thus attain ‘an entirely new conception of linguistic structure’ and ‘develop a certain fairly complex but reasonably natural algebra of transformations having properties we apparently require for grammatical description’ (SS 44) (cf. 2.82; 3.72f; 5.27, 86; 6.8, 29, 51, 60; 7.40, 7¹⁸). This ‘transformational model’ is still ‘more powerful’, applying to ‘languages beyond the bounds of phrase structure’ and ‘accounting’ for ‘relations’ between sentences’ ‘in a natural way’ (SS 6, 47, 75). Here, ‘apparently arbitrary distinctions’ are shown to ‘have a clear structural origin’ and to belong to a ‘higher-level regularity’ (SS 75, 107). ‘A wide variety of apparently distinct phenomena all fall into place’; ‘linguistic behaviour that seems unmotivated and inexplicable in terms of phrase structure appears simple and systematic’ from a ‘transformational point of view’ (SS 68, 75). Now, ‘apparently irregular behaviour’ and ‘glaring and distinct exceptions’ ‘result automatically from the simplest grammar’ and from ‘our rules’ for ‘regular cases’, and emerge as ‘instances of a deeper underlying regularity’ (SS 63f, 66f, 68, 85, 88f, AT 190). This is a ‘remarkable indication of the fundamental character of this analysis’ (SS 66f).

7.41 The transformational model must be ‘a generative grammar’: ‘a system of rules that in some explicit and well-defined way assigns structural descriptions to sentences’ (AT 8). Here, ‘the linguist's task’ is ‘to produce a device of some sort (called a grammar)’ for ‘generating all grammatical sequences’ and no ‘ungrammatical ones’ (SS 85, 11, 13). Hence, ‘the fundamental aim in the linguistic analysis of language is to separate’ these two sets and ‘to study the structure of the grammatical sequences’ (SS 13f). The ‘task’ may ‘explicate’ the ‘intuitive concept’ not just of ‘“grammatical in English”‘ but of ‘“grammatical”‘ in ‘general’ (SS 13). To simplify the task, a distinction is drawn between ‘grammaticalness’ in ‘competence’ versus ‘acceptability’ in ‘performance’ (AT 11) (cf. 7.12). Sentences may be ‘high’ on one ‘scale’ and ‘low’ on the other. ‘Grammaticalness is only one of many factors’ that ‘determine acceptability’, along with ‘memory limitations, intonational and stylistic factors, “iconic” elements of discourse, and so on’ (AT 10f) (cf. 7.13, 56).¹⁴ These limits may be probed with ‘operational tests’ of ‘rapidity, correctness, and uniformity of recall and recognition, or normalcy of intonation’, whereas ‘the much more abstract’ and ‘important notions of grammaticalness’ may not. It may in fact be ‘quite impossible to characterize’ ‘unacceptable sentences in grammatical terms’, e.g. by ‘formulating rules’ ‘to exclude them’ or by ‘limiting the number of reapplications of grammatical rules’ (AT 11f). For Chomsky's ‘purposes’, ‘“acceptable” refers to utterances that are perfectly natural and immediately comprehensible without paper-and-pencil analysis, and in no way bizarre or outlandish’ (AT 10).

7.42 Chomsky's reasoning leaves it unclear just how ‘a study of performance’ might ‘investigate the acceptability’ of ‘sentences’ or how a grammar could test its claims about ‘speakers’ ‘rejecting’ ‘sequences as not belonging to the language’ (cf. AT 12, SS 23). He postpones the problem by relying on ‘clear cases’ (preferably ‘violating purely syntactic rules’ rather than ‘semantic or “pragmatic”‘ ones), by not ‘appealing’ to ‘farfetched contexts’, and by promising that for ‘intermediate cases’ we can ‘let the grammar itself decide’ once it is ‘set up in the simplest way’ to ‘include clear sentences and exclude clear non-sentences’ (SS 14 16, AT 76, 208) (cf. 4.67; 7.21; 13.40). But such a stage presupposes that ‘linguistic theory’ has ‘stated the relation between the set of observed sentences and the set of grammatical sentences’ (SS 14n, 55), which has proven far harder than Chomsky implies here. He expects ‘a systematic account of how application of the devices and methods appropriate to unequivocal cases can be extended and deepened’ for others (AT 77f). Meanwhile, he follows his own judgment in adducing sentences that ‘we do not have’, or that ‘many would question the grammaticalness of’, or that are ‘much less natural’ (SS 73, 35n).¹⁵ Or, he proposes ‘degrees of grammaticalness’, depending on how ‘completely we violate constituent structure’, and hopes this ‘can be developed in purely formal terms’ (SS 36n, 43nf) (cf. 7.72, 93; 13.40). Doubtful cases would be ‘interpreted’ ‘by analogies to nondeviant sentences’, at least by ‘an ideal listener’ (AT 76, 78, 149). The snappiest evasion of all, which Chomsky admits leaves ‘a serious gap in theory’, is to ‘assume that the set of sentences is somehow given in advance’ (SS 103, 85, 11, 18, 54).

7.43 These issues bear on how a ‘theory’ might ‘provide a general method for selecting a grammar, given a corpus of sentences’ (SS 11). Noting that the whole ‘set of sentences’ a language allows for ‘cannot be identified with any partial corpus obtained by the linguist in his field work’, which is necessarily ‘finite and somewhat accidental’ (13.43), Chomsky defines ‘a language’ as ‘an infinite set of sentences,¹⁶ each finite in length and constructed out of a finite set of elements’ (SS 15, 13). ‘All natural languages in their spoken or written form are languages in this sense, since each’ ‘has a finite number of phonemes or letters’, and ‘each sentence is representable as a finite sequence’ of these (even a ‘formal system of mathematics’ qualifies) (SS 13) (cf. 6.51, 56, 60). ‘The assumption that languages are infinite’ can ‘simplify the description’ by allowing the ‘grammar’ to have ‘recursive devices’ for ‘producing infinitely many sentences’ (SS 23f). ‘Hence, a generative grammar must be a system of rules that can iterate to generate an indefinitely large number of structures’ (AT 16).

7.44 Specifically, ‘the infinite generative capacity of the grammar arises from a formal property’ of ‘rules’: ‘inserting’ basic structures ‘into others’, ‘this process being iterable without limit’ (AT 142). ‘In this respect, a grammar mirrors the behaviour of the speaker who’ ‘can produce and understand an infinite number of new sentences’, and ‘knowledge of a language involves this implicit ability’ (SS 15, AT 15). This notion is said to capture the ‘creative aspect’ ‘all languages have’: they ‘provide the means for expressing indefinitely many thoughts and for reacting appropriately to an indefinite range of new situations’ (AT v, 5) (cf. 3.38; 7.67, 83; 8.18, 28, 43, 83; 12.56, 58). The ‘technical devices for expressing a system of recursive processes’, as developed in ‘mathematics’, not only account for ‘sequences’ ‘longer than any ever before produced’, but offer ‘an explicit formulation of creative processes’ (AT 8, SS 17n ). Yet saying a language (or a grammar) is ‘creative’ because it has no longest sentence is like saying mathematics is creative because it has no highest number. Iteration and recursion are the exact opposite of creative: they just churn out the same thing at fixed increments. The real creativity of language, as shown, say, in poetry, would fall in a trouble zone of Chomsky's theory, namely on the borders of the ‘grammatical’, and unsettle his constructs (‘rules’, ‘restrictions’, etc.) for stating what can or cannot be ‘generated’ (cf. 13.40f).

7.45 Because ‘the grammar of a language is a complex system with many and varied interconnections between its parts’, ‘the linguist's task’ should be ‘to describe language’ by ‘a universal system of levels of representation’ (cf. Chomsky 1955) (SS 60, 85, 11, AT 222) (cf. 4.71; 5.34f; 8.51f; 9.30; 11.16f, 35, 56; 12.82; 13.27). Indeed, SS saw ‘“linguistic level”‘ as ‘the central notion in linguistic theory’, a ‘level’ consisting of ‘descriptive devices’ ‘for the construction of grammars’ (SS 11). Borrowing (without acknowledgement) from structuralism, Chomsky proposes to ‘rebuild the vast complexity of the actual language more elegantly and systematically by extracting the contribution’ of ‘several linguistic levels, each’ ‘simple in itself’, although the higher ‘levels’ must attain ‘more powerful modes of linguistic description’ by ‘increasing complexity’ (SS 42n, 11).¹⁷ Perhaps ‘we can determine the adequacy of a linguistic theory by developing rigorously and precisely the form of grammar corresponding to a set of levels’ and applying them to ‘natural languages’ (SS 11). ‘Each level’ is in turn ‘a system’ having a ‘set’ (or ‘alphabet’) of ‘minimal elements’ (‘“primes”’), plus ‘relations’, ‘mappings’ to other ‘levels’, and ‘operations of concatination which form strings’ of ‘arbitrary finite length’ (AT 222, SS 109).¹⁸

7.46 ‘A hierarchy of linguistic levels’ in ‘a satisfactory grammar of English’ might subsume ‘phonetic, phonological, word, morphological, phrase structure’, and ‘transformational structure’ (AT 223, SS 11f, 18, 85).¹⁹ Chomsky dislikes ‘the idea that higher levels are literally constructed out of lower level elements’, though he still depicts a ‘sentence’ as a ‘sequence’ of ‘phonemes, morphemes, or words’ (SS 58f, 106, 18, 32f). Against earlier American linguists, he sees ‘no objection to mixing levels’ (SS 57, 59, 106) (cf. 5.35; 13.27). The ‘interdependence of levels’ -- for instance, ‘if morphemes are defined in terms of phonemes’ and ‘morphological considerations’ are ‘relevant to phonemic analysis’ -- does not entrain ‘linguistic theory’ in any ‘real circularity’; the ‘compatibility’ of ‘phonemes and morphemes’ helps ‘lead to the simplest grammar’ (SS 57) (cf. 4.50; 5.36, 45; 13.27). This insight, however, ‘does not tell how to find the phonemes and morphemes in a direct, mechanical way’, but then ‘no other’ ‘theory’ does ‘either, and there is little reason to believe’ any can. This argument is meant to refute the ‘commonly held view’ that ‘syntactic theory is premature’ when ‘problems’ ‘on the lower levels’ are ‘unsolved’ -- ‘a faulty analogy between the order of development of linguistic theory and the presumed order of operations in discovery’ (SS 59f). ‘Developing one part of grammar’ is rather aided by a ‘picture’ of ‘a completed system’ (cf. 5.28; 13.43); and working down from ‘higher levels’ can bypass ‘absurd’ and ‘futile tasks’ like ‘stating principles of sentence construction in terms of phonemes and morphemes’ (cf. 7³³).

7.47 On ‘the level of phrase structure’, ‘each sentence is represented by a set of strings, not by a single string’ or ‘sequence’ as ‘on the level of phonemes, morphemes, or words’ (SS 32f, 47). This factor makes ‘phrase structure’ a ‘level’ of ‘different and nontrivial characteristics’, not able to be ‘subdivided’ into some ‘set of levels’ ‘ordered from higher to lower’ (SS 32f, 47). ‘The break between’ ‘phrase structure and the lower levels’ is ‘not arbitrary’, because ‘similarities’ and ‘dissimilarities’ of ‘representation’, such as ‘rules with ‘different formal proper

7.48 In SS, Chomsky resolves the levels into ‘a picture of grammars’ with ‘a natural tripartite arrangement’, each part having its own set of ‘rules’: ‘phrase structure’, ‘transformational structure’, and ‘morphophonemics’ (SS 45f, 114).²⁰ ‘A grammar’ provides ‘rules’ for ‘reconstructing’ ‘phrase structure’ and for ‘converting strings of morphemes into strings of phonemes’; these two rule sets are ‘connected’ by ‘a sequence of transformational rules’ for ‘carrying strings with phrase structure into new strings to which morphophonemic rules can apply’ (SS 107). ‘Phrase structure’ is created by means of ‘instruction formulas’ called ‘rewriting rules’ (SS 26, 29, 110, AT 66). For example, , ‘X --> Y is interpreted: “rewrite X as Y”‘, or ‘A --> Z/X => Y’ as ‘category A is realized as the string Z’ ‘in the environment’ ‘of X to the left and Y to the right’; in ‘context free rules’, ‘X and Y are null, so that the rules apply independently of context’ (SS 26, 110, AT 66) (cf. 7.73f; 13.39).²¹ These ‘rules’ apply to ‘only a single element’ and proceed in a ‘sequence’ from an ‘initial symbol’ to ‘a terminal string’, where ‘no further rewriting is possible’ (SS 29f, 26f, AT 66f).²² The sequence of strings is termed a ‘derivation’, and each sentence has a ‘history of derivation’ (SS 37, 107) (cf. 7.52). A ‘system of rewriting rules’ can ‘present’ ‘grammatical information’ ‘in the most natural way’, state ‘grammatical functions’, and ‘provide a simple method for assigning a unique and appropriate phrase marker to a terminal string, given its derivation’ (AT 72, 66f).

7.49 For Chomsky, ‘the most obvious formal property of utterances is their bracketing into constituents’ (AT 12) (cf. 3.22; 4.59; 5.21, 50, 62; 7.63; 9.33; 13.26). Hence, ‘constitutent analysis’ by ‘parsing’ is ‘customary’ for ‘linguistic description on the syntactic level’ (SS 26) (cf. 11.3, 14, 16, 33f, 77, 79, 94). As a better means for ‘assigning’ or ‘imposing constituent structure’ and obtaining ‘valuable, even compelling evidence’ about it, Chomsky proposes ‘transformations’ (SS 83, 73, 81). ‘A grammatical transformation’ is ‘a rule’ that ‘applies to a string with a particular structural description’ (AT 89). So ‘a transformation is defined by the structural analysis of the strings’ it ‘applies’ to and by ‘the change it effects’ on them (SS 111, 61, 91). Reciprocally, ‘the representation of a string on the level of transformations is given by the terminal string’ and the ‘transformations by which it is derived’ (SS 91).

7.50 A ‘transformational treatment’, as we saw, is claimed to enhance ‘simplicity’ (SS 62, 72, 81) (cf. 7.36ff). As a ‘general principle’ stressed in SS, ‘if we have a transformation that simplifies the grammar and leads from sentences to sentences in a large number of cases’, we can ‘simplify the grammar even further’ by ‘assigning constituent structure to sentences in such a way that this transformation always leads to grammatical sentences’ (SS 83). And we can gain still more by using ‘the same underlying transformational pattern’ for several constructions, e.g. ‘negatives and interrogatives’ (SS 64ff). But in AT, Chomsky says these ‘operations’ need no ‘a priori justification’ by being ‘the most simple or “elementary” ones’ (AT 55). ‘“Elementary operations”‘ are often not ‘transformations at all, while those that are may be ‘far from elementary’. ‘Transformations’ ‘manipulate substrings only in terms of their assignment to categories’, ‘independently’ of the ‘length or internal complexity of strings belonging to these categories’. Evidently, Chomsky's hopes for simplicity were waning (cf. 7.36).

7.51 In SS, however, he had proposed to ‘simplify the description of English and gain new insight’ by ‘limiting the direct description in terms of phrase structure to a kernel of basic sentences: simple, declarative active with no complex verb or noun phrases’ (SS 106f, 47f, 61) (cf. 4.58, 65). ‘All other sentences can be described more simply as transforms’, i.e. ‘strings’ ‘derived’ by ‘simply statable transformations’ (SS 80, 48). Chomsky promised ‘clear and easily generalizable considerations of simplicity’ for ‘determining which set of sentences belongs to the kernel, and what sorts of transformations’ can ‘account for the non-kernel sentences’ (SS 77). We thus establish ‘the transformational level'²³ where ‘an utterance is represented’ ‘abstractly’ via ‘a sequence of transformations by which it is derived, ultimately from kernel sentences’ (SS 47). The ‘kernel of the language’ can be ‘defined’ ‘as the set of sentences produced by obligatory transformations’ (SS 45f, 61, 91), i.e., those needed to make a sentence at all. All others pass through a ‘transformational history’, which might be a concept for ‘determining sentence types in ‘general’ (SS 71, 89, 91f, AT 130) (cf. 7.48; 13.28).

7.52 Neither SS nor AT remotely attempts to give a comprehensive list of transformations. A handful are cited for demonstration, such as those for ‘negations’, ‘questions’, ‘emphatic affirmatives’, ‘nominalizations’, and ‘comparatives’ (SS 61f, 64, 66, 72, 111-14, AT 178). When ‘adjective modifiers’ are derived from whole sentences (‘converting “the boy is tall” into “the tall boy”’) or ‘relative clauses’ (i.e. ‘“the boy who is tall”’) (SS 72 AT 217), we see simpler constructions being derived from more complex ones, not the other way around (cf. 13.54). And these very cases are adduced as ‘independent support’ and ‘syntactic justification for a transformational analysis’ (SS 73, AT 189).

7.53 A ‘paradigmatic instance’ is ‘the passive transformation’ (SS 77). Having ‘both actives and passives in the kernel’ would make ‘the grammar’ ‘much more complex’ ‘than if passives are deleted and reintroduced by a transformation’ for ‘inverting subject and object’, and, where needed, ‘deleting the unspecified agent’ (SS 77, 72, 79; AT 128) (cf. 9.70). ‘“Quantificational” sentences such as ‘“everyone in the room knows at least two languages”‘ versus ‘“at least two languages are known by everyone in the room”‘ lead to the overstatement that ‘not even the weakest semantic relation (factual equivalence) holds in general between active and passive’, whose ‘relations’ would ‘not have come to light’ in terms of ‘synonymy’ -- a warning against ‘vague semantic clues’ (SS 100f).²⁴ But the relation, I think, is semantic similarity plus difference in focus: in each sentence, focus goes to the identity and size of the set named in subject position (just which persons versus just which languages), while the set named via direct object or prepositional agent is not in focus (any two languages versus anyone in the room).

7.54 In SS Chomsky felt that despite possible ‘difficulties and complications’, ‘the order’ must be defined among the ‘rules’ and ‘transformations’ (SS 32, 111, 114, 66, 44, 35). But in AT he sees ‘no way’ to decide ‘a priori’ whether ‘the rules’ should be ‘ordered’ or ‘unordered’, e.g., in terms of ‘simplicity’ or ‘elegance’ (AT 39). And ‘the theory of transformation markers permits a great deal of latitude’ in ‘the ordering of transformations’ (AT 133).²⁵ Yet Chomsky evokes ‘strong reasons’ for ‘assigning an internal organization and an inherent order of derivation’ (AT 125).²⁶ He uses the term ‘sequential derivations’ for those ‘preserving ordering’ and considers them central to ‘empirical studies of transformational grammar’ (AT 67, 211).

7.55 Chomsky limits his ‘theory of transformations’ by not covering ‘the full range of possibilities for stylistic inversion’ and ‘reordering’ (AT 126f, 227f). ‘Richly inflected languages’ (e.g. Russian, Mohawk) ‘tolerate’ much ‘reordering’ unless it ‘leads to ambiguity’ -- maybe a ‘universal’ constraint (AT 126f) (cf. 3.27). Again saving labour while advancing his cause, Chomsky insists that ‘the rules of stylistic reordering are very different from grammatical transformations’, the latter being ‘much more deeply embedded in the grammatical system’ and using ‘markers drawn from a fixed, universal language-independent set’, while ‘stylistic’ ones are ‘peripheral’, apply to ‘performance’, and ‘have no apparent bearing’ ‘on the theory of grammatical structure’ (AT 127, 222f).

7.56 In contrast to SS, AT presents ‘grammar’ with a different group of three parts: ‘a syntactic’, ‘a semantic’, and ‘a phonological component’ (AT 141, 16) (cf. 7.48; 9.30). Morphemics fades, and semantics finally enters the picture. The exact ‘relation between syntax and semantics’ is a question Chomsky often raises and fails to resolve, regarding it as ‘a side issue’ and a ‘dangerous ground’ fraught with ‘difficulties’, ‘confusion’, and ‘speculation’ (SS 93, 101n, AT 75, 148). But he devotes explicit sections to it in both SS (92-105) and AT (14-163). In SS, he follows structuralists like his teacher Harris (1951) by asserting ‘the independence of grammar’ from ‘semantics’ and ‘concluding that only a purely formal basis can provide a firm and productive foundation for construction of grammatical theory’ (SS 13, 106, 100) (cf. 5.61; 13.54). His own ‘theory’ is ‘completely formal and non-semantic’, ‘meaning’ being no more relevant for ‘constructing a grammar’ than ‘the hair colour of speakers’ (SS 93). The idea that ‘one can construct a grammar with appeal to meaning’ is ‘widely accepted’ (cf. 3.40) but ‘totally unsupported’, and obscures ‘important generalizations about linguistic structure’ (SS 93, 101, AT 32, 78).

7.57 So SS proffers ‘a purely negative discussion of the possibility of finding semantic foundations for syntactic theory’, leading to the ‘conclusion that grammar is autonomous and independent of meaning’ (SS 93, 17). Chomsky denies having ‘acquaintance with any detailed attempt to develop a theory of grammatical structure in partially semantic terms’ (though the traditional grammars he praises in AT were just this, 8.5, 12.41) nor any ‘rigorous proposal for using semantic information in constructing or evaluating grammars’. He turns the tables by placing ‘the burden of proof’ ‘completely on linguists who claim’ to ‘develop some grammatical notion in semantic terms’ (SS 94)²⁷ -- a bold move, since the burden of proof should rest on the challenger of ‘widely accepted’ views. Or, he advocates postponement, because ‘a decision as to the boundary separating syntax and semantics’ ‘is not a prerequisite for theoretical and descriptive study’ (AT 160). ‘Correspondences’ and ‘relations between semantics and syntax can only be studied after the syntactic structure has been determined on independent grounds’ and within ‘some more general theory’ (SS 17n, 102, 108). Semantics can wait, but syntax can't (7.46).

7.58 Moreover, ‘despite the undeniable interest and importance of semantic’ ‘studies of language’, ‘semantic notions’ like ‘reference, significance, and synonymity’ ‘have no direct relevance’ for ‘characterizing the set of grammatical utterances’, which is the true ‘problem of syntactic research’ (SS 17, 102f).²⁸ His demonstration that ‘“grammatical” cannot be identified with “meaningful” or “significant”‘ hinges on the now-famous sentence ‘“colourless green ideas sleep furiously”‘ (SS 15, AT 149). This example reminds me of the grammatical but nonsensical examples adduced by structuralists to show the self-sufficiency of ‘grammatical categories’, like Chomsky's own ‘“Pirots karulize etalically”‘; but he finds their ‘notion of “structural meaning”‘ ‘quite suspect’ and doubts ‘that the grammatical devices’ are ‘used consistently enough so that meaning can be assigned to them’ (SS 104, 108, cf. SS 103n, 104f) (cf. 5.61).²⁹

7.59 Chomsky favours moving in the reverse direction: ‘linguists with a serious interest in semantics’ should ‘deepen and extend syntactic analysis’ to supplant ‘unanalysed semantic intuition’ (AT 75) (cf. 7⁵). A ‘purely formal investigation’ of ‘syntactic structure’ ‘isolated and exhibited by the grammar’ is ‘rated more highly’ when it ‘provides’ ‘insight into problems of meaning and understanding’ and ‘supports semantic description’ (SS 93, 102, 12). He believes his own ‘requirement’ of a ‘completely formal discipline is perfectly compatible’ for making ‘significant interconnections with a parallel semantic theory’ (SS 103). However ‘imperfect’ and ‘inexact’, the ‘striking correspondences’ between ‘syntactic structures’ ‘in formal grammatical analysis and specific semantic functions’ at least indicate that ‘syntactic devices’ are ‘used fairly systematically’ (SS 104, 101f, 108).

7.60 Such arguments help us to appreciate why AT adds a ‘semantic component’ and states that ‘a descriptively adequate grammar’ is also ‘a theory of sentence interpretation’ (AT 76). Yet Chomsky hedges his bets by requiring ‘that all information utilized in semantic interpretation must be presented in the syntactic component’ (AT 75). At one point, he unexpectedly says he sees ‘no way to decide’ ‘a priori’ ‘whether the burden of presentation should fall on the syntactic or semantic component’; ‘in fact, it should not be taken for granted that syntactic and semantic considerations can be sharply distinguished’ (AT 77f). But this indecision applies specifically to an issue he is reluctant to face anyway, namely ‘deviant sentences’. ‘The syntactic component’ could only treat them via ‘structural similarities to perfectly well-formed’ ones, while the ‘semantic’ one could provide ‘lexical items’ to ‘specify’ ‘rules’ for ‘determining the incongruity’ (cf. 7.72). Chomsky proposes ‘selectional rules’ that ‘impose a hierarchy of deviation from grammaticalness’ on ‘sentences’ ‘generated by relaxing’ ‘constraints while keeping the grammar unchanged’ (AT 152). These rules could be ‘dropped from the syntax’ and put in ‘the semantic component’ with ‘little violence to structure of the grammar’ (AT 153, 158).

7.61 In practice, the role of meaning is marginalized by sticking close to the literal wording of sentences or by replacing words with symbols, and suggesting thereby that the relations between ‘transforms’ depend on the identities of written words or symbols rather than on similarities of meaning (cf. 13.33). Emblematic here Chomsky's use of ‘ambiguity and similarity of understanding’ as ‘a motivation for’ the ‘levels’ of ‘phrase structures’ and ‘transformations’, e.g., to clear up constructions like ‘“the shooting of the hunters”‘ or ‘“flying planes can be dangerous”‘ (SS 86ff, AT 21). ‘Transformational grammar’ offers a way to make ‘ambiguity’ ‘analysable in syntactic terms’ by treating it as ‘constructional homonymity’ (SS 28, 33, 86fn, 107). An ‘ambiguity’ signalled by some ‘intuitive similarity of utterances’ points to ‘dual representations on some level’, while ‘synonymity’ points to a ‘similar or identical representation’ (SS 107). A very simple case is ‘a phoneme sequence’ that can be ‘ambiguously understood’ in terms of ‘morphemes’ (e.g. /neym/ as ‘“a name” or “an aim”’) (SS 85) (a prominent trait of English).

7.62 But the most elaborate and explicit means for absorbing semantics and meaning into ‘the syntactic component’ is by adding the conception of a ‘deep structure’ that ‘underlies’ the ‘actual sentence’ and ‘determines its semantic interpretation’ (AT 16, 23, 135, 138). This ‘deep structure’ ‘expresses’ ‘grammatical relations’ and gets ‘mapped’ by ‘transformational rules’ into ‘surface structure’ (AT 99, 16f, 123, 99, 135). Chomsky complains that the ‘syntactic theories’ in ‘modern structural (taxonomic) linguistics’ ‘assume that deep and surface structures are actually the same’ (AT 16) (cf. 7.80). Making the two ‘distinct’ is now ‘the central idea of transformational grammar’, with ‘surface structure determined by repeated application of formal operations called “grammatical transformations” to objects of a more elementary sort’ (AT 16f) (which are also more abstract than ‘kernels’ and look less like sentences). Against ‘immediate constituent analysis’, which ‘may be adequate’ for the ‘surface’, Chomsky's ‘concern’ is ‘primarily with deep structure’ as the site of ‘the significant grammatical relations of an actual sentence’, while he finds ‘surface structure’ ‘unrevealing’ and ‘irrelevant’ (AT 17, 220, 24) (7.84, 82; cf. 13.47). He warns how ‘surface similarities hide underlying distinctions’ which ‘no English grammar has pointed out’, nor are they ‘clear to the speaker’ (AT 24, 22) (cf. 13.49). ‘Universality is claimed’ only for ‘deep structures’, whereas ‘surface structures’ need have no ‘uniformity’, as ‘modern linguistics’ has indeed ‘found’; this argument keeps their ‘findings’ from conflicting with ‘the hypotheses of universal grammarians’ (AT 118) (cf. 7.18f, 55).³⁰ Also, ‘a very general, and perhaps universal way’ for ‘defining the system of grammatical relations’ is ascribed to the ‘categorial component’, which ‘determines the ordering of elements in deep structures’ (AT 122f).

7.63 This new design enables ‘grammatical functions’, which are ‘relational’ and sometimes have ‘traditional’ ‘names’ like ‘“Subject” and “Object”‘, to be ‘sharply distinguished from ‘grammatical categories’ like ‘“Noun Phrase”‘ and ‘“Verb Phrase'“ (AT 68f, 104) (cf. 4.69; 5.55, 5³⁰; 6.49; 9.46, 70; 11.34; 12.70, 79).³¹ ‘The so-called “grammatical Subject” belongs to ‘surface structure’, while ‘functional notions’ like ‘“logical Subject”‘ belong to ‘deep structure’; thus, a ‘“grammatical Subject” may be a ‘“logical Object”‘ (AT 70, 23, 68, 163, 230). This ‘difference’ ‘provides the primary motivation and empirical justification for the theory of transformational grammar’ (AT 70) (cf. 7.79). ‘A sentence’ is assumed to have ‘a basis’ whose ‘phrase markers’ ‘directly represent’ ‘semantically relevant information concerning grammatical function’ (AT 70, 117, 230). Yet ‘the extension to surface structure of such functional notions’ is not ‘entirely straightforward’, and ‘different definitions are needed’ (AT 220f). For example, ‘Topic-Comment’ might be ‘the basic grammatical relation of surface structure corresponding (roughly) to the fundamental Subject-Predicate relation of deep structure; ‘often’ ‘Topic and Subject coincide’. Disregarding communicative functions, Chomsky ‘defines the Topic’ ‘as the leftmost NP immediately dominated by S [Sentence] in surface structure and the Comment’ as ‘the rest of the string’ (but cf. 9.57, 9²⁷; 11.45, 64). Also, ‘case’ is attributed to ‘the position of the Noun in surface structure rather than deep’ (AT 221f), an idea ‘case grammar’ was later to dispute (cf. 11.28, 41).

7.64 The home of ‘underlying’ or ‘deep structure’ is ‘the base of the syntactic component’ (AT 17). This is ‘a system of rules that generate a highly restricted (perhaps finite) set of basic strings, each with an associated structural description called a base Phrase marker’ -- the ‘elementary unit’ of ‘deep structure’ (AT 17, 135). ‘The basis of the sentence’, namely its ‘underlying’ ‘sequence of base phrase markers’, gets ‘mapped into the sentence by the transformational rules’, which ‘automatically assign to the sentence a derived phrase marker, ultimately, a surface structure’ (AT 17, 128, 135). To ‘simplify the exposition’, Chomsky assumes that ‘no ambiguities are introduced by rules of the base’ (AT 17). The ‘“kernel sentences”‘ of SS are briefly mentioned as having ‘a single phrase marker’ and ‘involving a minimum of transformational apparatus’; but they have a merely ‘intuitive significance’ (AT 18) (cf. 7.51, 53, 82).

7.65 Being associated with ‘deep structure’ and ‘the categorial component’, ‘much of the structure of the base’ ‘is common to all languages’ (AT 123, 117) (cf. 7.62). Chomsky finds it ‘natural to suppose that formal properties of the base will provide the framework for the characterization of universal categories’: either the ‘rules’ themselves, or ‘constraints’ on the choice’ of ‘rules’ and ‘elementary structures’, or ‘a fixed’ ‘alphabet’ of ‘category symbols’ (AT 117, 141f). And ‘the form of the categorial component’ may be ‘determined by universal conditions that define “human language”‘ (AT 120f). This ‘traditional view’ dates back ‘at least to the Grammaire generale et raisonnee’ and in light of ‘relevant evidence’ remains ‘true’ ‘today’ -- and convenient too: ‘aspects of base structure’ ‘not specific to a particular language’ ‘need not be stated in the grammar’ (AT 117) (cf. 7.19f, 72).

7.66 Therefore, ‘the account of the base rules’ ‘may not belong to the grammar of English any more than the definition’ ‘of “transformation”‘ (AT 117f). ‘Not all phrase markers generated by the base will underlie actual sentences and thus qualify as deep structures’ (AT 138). But ‘the transformational rules provide’ a ‘very simple’ ‘test’, namely whether the ‘rules can generate’ a proper ‘surface structure’ from a ‘deep’ one. This ‘filtering function’ of the ‘rules’ was ‘true of the earlier version’ (in SS) as well though ‘never discussed’ (AT 191, 139), since ‘deep structure’ hadn't been proposed then. Transformations which didn't filter would be useless, since they wouldn't supply the restrictions the formalism requires.

7.67 ‘The syntactic component’ is the ‘sole “creative” part’ of the grammar’, while the other ‘two are purely interpretive’ and ‘play no part’ in ‘generating sentence structures’ (AT 135f, 141, 16) (cf. 7.44; 13.32). ‘The deep structure of a sentence is submitted to the semantic component for semantic interpretation, and its surface structure enters the phonological component’ for ‘phonetic interpretation’ (AT 135, 141, 16, 138, 99).³² ‘The final effect of a grammar, then, is to relate a semantic interpretation to a phonetic’ one -- ‘that is, to state how a sentence is interpreted’ (AT 136). ‘When we define deep structures’ as ‘“generated by the base”‘, we ‘assume that the semantic interpretation of a sentence depends only on its lexical items and the grammatical functions and relations represented in the underlying structures’. ‘This is the basic idea that has motivated the theory of transformational grammar since its inception’, though not ‘clearly formulated’ until Katz and Fodor (1963) (cf. 5.76; 7.71).³³

7.68 Though they are ‘not our concern’ and have not been ‘worked out’, Chomsky thinks his two ‘interpretive components function in parallel ways’ (AT 143). ‘Phonological rules’ ‘apply’ ‘first to minimal elements (formatives), then to the constituents of which they are parts’, ‘and so on, until the maximal domain of phonological processes is reached’. ‘In this way a phonetic representation of the entire sentence is formed’ via ‘the intrinsic abstract’ ‘properties of its formatives’ and ‘categories’ ‘in the surface structure’, and ‘related’ to a ‘signal’ (AT 143f, 16). ‘Similarly, the projection rules of the semantic component operate on the deep structure’ and ‘assign’ ‘a “reading” to each constituent on the basis’ of those ‘assigned to its parts’, all the way down to ‘the intrinsic semantic properties of the formatives and categories’ ‘in the deep structure’ (AT 144) (cf. 7.82; 13.59).

7.69 Chomsky's ‘syntactic component’ centred between the other two, is naturally the one AT is ‘concerned with’ (AT 3), and is assigned a whole range diverse and complex duties. It provides ‘rules that specify the well-formed strings of minimal syntactically functioning units (“formatives”) and assigns structural information to these strings and strings which deviate from well-formedness’. It ‘specifies the infinite set of abstract formal objects, each of which incorporates all the information relevant to a single interpretation of a particular sentence’ (AT 16). It ‘contains a lexicon’, with each ‘item specified’ by its ‘intrinsic’ ‘features’, such as ‘Animate’ and ‘Human’ (AT 78, 82f, 85f, 150f, 153, 226) (cf. 4.69; 7.72, 74).³⁴ Though they are evidently borrowed over from reality, Chomsky insists these features belong to ‘purely syntactic rules’ and to the ‘syntactic component, no matter how narrowly syntax is conceived’ (AT 150f). Maybe their absence in traditional grammar worries him.

7.70 The ‘lexicon’ is designed to be not just a list of ‘dictionary definitions’ (AT 37), but an enormously diverse and complex complement for the grammar (cf. 13.52). It contains not words as such, but ‘formatives’: ‘minimal syntactically functioning units’, including both ‘lexical items (“sincerity, boy”) and grammatical items (“Perfect, Possessive”)’ (AT 87, 3, 65) (cf. 13.30). Since ‘a careful grammar’ ‘reveals that many formatives have unique or almost unique grammatical characteristics’, ‘substantial simplification’ is gained (and labour saved) by putting all these ‘properties’ into ‘the lexicon’, ‘where they most naturally belong’, and ‘excluding’ them ‘from the rewriting rules’ (AT 86f) (cf. 13.59). We thus have the ‘advantage’ of making ‘the lexical entries’ absorb all ‘idiosyncrasies’ and ‘irregularities of the language’ (AT 86f, 216, 142, 214) (cf. 4.49, 52; 13.59; Sweet 1931:31).

7.71 The ‘entries’ in ‘the lexicon’ ‘each consist of a descriptive feature matrix and a complex symbol’ (AT 164, 222). The ‘features’ are ‘phonological’, ‘syntactic’, and, in the ‘definition’, ‘purely semantic’ (cf. Katz & Fodor 1963) (AT 214). The ‘semantic features’ are ‘distinct but related’ to the ‘syntactic’ ones and ‘constitute a well-defined set in a given grammar: a feature belongs’ here if ‘it is not referred to by any rule of the phonological or syntactic component’ (AT 88, 120, 142). Lest he imply that ‘the system of dictionary definitions’ is without ‘intrinsic structure’ and hence ‘atomistic’, Chomsky invokes ‘relations of meaning (rather than relations of fact)’ that ‘cannot in any natural way be described within’ ‘independent lexical entries’ (AT 160). Examples include the ‘“field properties”‘ studied in ‘componential analysis’, or the ‘referential domains’ of terms that are ‘mutually exclusive’, e.g. ‘colour words’ (AT 229, 160f, 229) (cf. 4.22; 5.68; 7.32). Moreover, Chomsky foresees ‘universal language-independent constraints on semantic features’ -- ‘in traditional terms, the system of possible concepts’ (AT 160) -- presumably innate rather than extracted from reality (but cf. 7.69; 13.24). This hope again rests on a parallel to ‘the theory of phonetic distinctive features’, which gives ‘a language-independent significance to the choice of symbol’ for ‘lexical formatives’ (cf. 7.20; 13.26); Chomsky ‘assumes’ that ‘the grammatical formatives and the category symbols’ in the lexicon ‘too are selected from a fixed universal vocabulary, although this assumption’ has no ‘effect’ on his ‘descriptive material’ (AT 65f, 160). Only ‘our ignorance of the relevant psychological and physiological facts’ hides the ‘a priori structure’ in ‘the system of “attainable concepts”‘ (AT 160).

7.72 The ‘lexical entry’ gets the tough jobs of supplying ‘information’ for ‘the phonological and semantic components’ and for ‘the transformational part of the syntactic component’, and of ‘determining the proper placement’ ‘in sentences, and hence by implication degree and manner of deviation of strings’ not ‘generated’ (again relieving the ‘rules’) (AT 87ff). Some ‘features’ are ‘inherent’ to a ‘formative’, being ‘part of the complex symbol of the lexical entry’, while others are ‘contextual’, being ‘introduced by grammatical rules’ when the ‘item’ is ‘inserted into a phrase marker’ (AT 171, 122, 176f). ‘This analysis of a formative as a pair of sets of features’ is ‘tentatively proposed’ ‘as a linguistic universal’ on ‘slender evidence’, but with the (labour-saving) corollary that the ‘features need not actually be mentioned’ ‘in the rules of the grammar’ (AT 181, 141f) (cf. 7.19, 65). Also, a possible explosion of features in the lexicon is to be counteracted by ‘redundancy rules’ for ‘ predictable’ ‘feature specifications’ -- another carry-over from the ‘phonological’ domain to the ‘syntactic’ and a prompt new candidate for ‘universal notation’ (e.g. ‘[+ Human]’ again) (AT 168f, 222, 166).

7.73 ‘A lexical entry is substituted’ in ‘positions’ where ‘its contextual features match those of the symbol’ (AT 121). Since the ‘rules’ can ‘assign’ ‘contextual features’ also to the ‘frames in which a symbol appears’, we have ‘context-sensitive rules’ in the lexicon instead of ‘context-free’, and thereby get them too out of ‘the grammar’ (AT 121f) (cf. 7.48; 13.52). In exchange, the ‘features determined by context’ are ‘noninherent’ and, being ‘unspecified in underlying structures’ and ‘added’ by ‘redundancy rules’, can make ‘no contribution to sentence interpretation’ (AT 182) (cf. 7.82). Or, ‘determining the restrictions’ via ‘feature specifications’ might be ‘preferable, since it does not affect the structure of the lexicon’, leaving it ‘simply a list of entries’ (AT 188). Chomsky wonders whether to ‘list in the lexicon only the features’ for ‘frames in which the item’ ‘cannot appear’; or to list those in which it ‘can appear’ and to assume it is ‘automatically’ ‘specified negatively’ for ‘every contextual feature not mentioned’ (AT 110f, 166, 230). Or, a ‘distinct subpart of the base of the syntactic component’ could contain ‘lexical rules’ derived by a ‘general analysis of lexical items’ and empowered to ‘introduce lexical formatives’ (AT 68, 190, 74).

7.74 Beyond ‘the lexicon’, Chomsky's ‘base’ has a ‘categorial component’, ‘defined’ as a ‘system of rewriting rules’ (AT 123, 120) (cf. 7.48). ‘Among the rewriting rules’ ‘we can distinguish branching rules’, which ‘analyse a category symbol’ ‘into a string of (one or more) symbols’, either ‘terminal’ or ‘nonterminal’; versus ‘subcategorization rules’, which ‘introduce syntactic features and thus form or extend a complex symbol’ (AT 112). Again, these two types of ‘rule may be context-free’, ‘introducing inherent features’ (like ‘[Human]’); or ‘context-sensitive’, ‘introducing contextual features’ (like ‘[Transitive]’) by ‘environment’ (AT 112, 120, 90, i.r.).³⁵ In ‘theory’, ‘a grammar is obviously more highly valued if subcategorization is determined by a set of contexts that is syntactically definable’ (AT 97f). But midway through AT, Chomsky decides it would be less ‘restrictive’ and more ‘flexible’ to change ‘the base component’ in his ‘earlier proposal’ by discarding ‘subcategorization rules’ as ‘rewriting rules and ‘assigning’ them too ‘to the lexicon’, thus leaving only ‘branching rules’, ‘all context-free’ (like in a ‘simple phrase structure grammar’), inside ‘the categorial component’ (AT 123, 120, 128, 122). Now, ‘strict subcategorial and selectional restrictions’ get ‘listed in lexical entries’ and ‘defined by transformational rules’ (AT 139).

7.75 The place of morphology in the tripartite AT model, lacking the older ‘morphophonemic level’ (7.48), is merely sketched. Chomsky ‘compares’ ‘the traditional method of paradigms’ whose ‘dimensions’ are ‘inflectional categories of gender, number, case, and declensional type’, against ‘the descriptivist method of morphemic analysis’, which Chomsky (after Hockett) classes as an ‘“item-and-arrangement” grammar’ (AT 170, 172, 232) (cf. 4.57; 5.50, 74, 5¹²; 6.34; 7.75f; 8.57; 9.31, 9¹¹; 12.71). The older ‘paradigmatic description’ can be ‘restated’ or ‘incorporated directly’ in ‘the theory of syntactic features’, using ‘not + and - but integers’ ‘associated with traditional designations’ (AT 171f).³⁶ The concept of ‘morphemes’ in ‘modern linguistics’, in contrast, is ‘clumsy’ and ‘inelegant’ to ‘represent’ by ‘a grammar based on rewriting rules or transformations’, and is ‘in fact designed’ ‘to exclude’ ‘all but the most elementary general rules’ (AT 173, 323). ‘The order of morphemes is quite arbitrary, whereas this arbitrariness is avoided in paradigmatic treatment, the features being unordered’ (AT 174). Also, ‘many morphemes are not phonetically realized’ and so are treated, ‘in particular contexts, as zero elements’, such that ‘each case’ needs ‘a specific context-sensitive rule’; Chomsky finds these rules ‘superfluous’, and his ‘feature analysis simply gives no rule’ (AT 173). So he judges ‘the modern descriptivist reanalysis’ ‘in terms of morpheme sequences’ ‘an ill-advised theoretical innovation’ (AT 174) (cf. 7.5).

7.76 But he doesn't quite throw it out: ‘within our framework, either paradigmatic analysis’ by ‘features or sequential morphemic analysis is available, whichever permits the optimal and most general statement’ (AT 174).³⁷ He foresees no ‘difficulty in extending the theory of transformations’ to ‘formalize traditional rules’ for ‘inflectional systems’ (AT 175f). Yet he admits that morphemic ‘derivation processes’ are a bigger ‘problem’ for ‘generative’ grammar because ‘they are typically sporadic and only quasi-productive’, even when ‘the meaning’ is ‘to some extent predictable’ from ‘inherent semantic properties of the morphemes’ (AT 184, 186). Still, ‘all presently known theories of language fail’ here, so ‘quasi-productive processes’ need not ‘support an alternative theory of grammar’ (AT 235f) (argued by Bolinger 1961).

7.77 Though I have had to pass over some technical details about ‘rules’, ‘features’, and so on, the major ‘aspects’ of Chomsky's ‘theory of syntax’ should now be evident. The two books I have summarized are far from being the whole story. He and his associates produced a flood of additional books and papers revising numerous aspects of the model or refuting objections to it; in fact, the grammar got more and more ‘transformational’ because it kept getting transformed, even to the point where many ‘transformations’ themselves were dropped, along with ‘deep structure’ (Chomsky 1977). But AT and SS were and still are by far the most widely read of his works, and Chomsky himself later (1971[1968]:184f) called AT ‘the standard theory’. I have treated the two books mainly as two advocacies of one overall project, while pointing out the important innovations and greater development of AT over SS. The vacuum left by not acknowledging his debt to ‘modern’ linguistics (7.5, 19, 29, 57, 76) is filled in AT by garnering sources among philosophers of previous centuries (7.3) and by citing himself as a main source. AT contains 113 references to 18 works he had by that time authored or co-authored,³⁸ and relies heavily on works of his associates and followers (e.g. Halle, Miller, Katz, Fodor, Postal, Lees, Bever).

7.78 Above all, AT reveals a marked change in Chomsky's estimation of the project. He now presents the approach not merely as a way to make English grammar ‘simpler’, more ‘adequate’, and so on, but as a leading candidate in the investigation of ‘linguistic competence’, ‘innate language capacity’, ‘language acquisition, ‘language universals’, and ‘philosophy of mind’. To retain much of the conceptions and terminology originally designed for ‘grammatical description’, he makes them ‘systematically ambiguous’, even though the correspondence between a speaker's or a child's knowledge and ‘the linguist's account’ is precisely what linguistics has to demonstrate (cf. 7.15, 28; 13.45). He thus lends a more exalted significance to the rather traditional centrepiece of his project: analysing, describing, and comparing phrases or sentences in terms of grammatical structure.

7.79 Since many grammatical relations and distinctions are not formally marked in English, Chomsky promises to offset the lack by postulating ‘underlying structures’, ‘markers’, and ‘features’. He stresses superfical similarity with underlying difference, and underlying similarity with superficial difference (cf. 7.14, 53, 61, 82; 13.7, 30, 43, 54). We are justified in ‘transforming’ linguistic data to put such cases into clearer forms, which competing ‘grammars’ fail to do. He nourishes the prospect that, on the ‘underlying’ (‘deep’, ‘base’, etc.) level, structure is far more precise, distinct, orderly, and general than everyday utterances indicate. Down there, categories are ‘universal’ (‘determined by the nature of language as such’), relationships are crystal-clear, constraints on what may or may not be constructed are exactly stated, meanings are rigorously decomposed, and ambiguities are fully resolved. Such demanding goals force Chomsky to be evasive about his deep level, or to advance peculiar claims. To limit ‘functional’ ‘issues’ to ‘surface structure’, he says ‘order’ ‘plays no role in determining grammatical relations in deep structures’, even though he has already said ‘the categorial component determines the ordering of elements in deep structures’ (AT 220f, 122ff). To keep his ‘deep structure’ simple, he says ‘features determined by context’, being ‘unspecified in underlying structure’, can make ‘no contribution to sentence interpretation’ (AT 182) (7.72); but what can ‘context’ be if not an influence on how people interpret?

7.80 Chomsky's evasions are predictably clustered at points of contact between his technical, underlying apparatus and the facts or patterns it purports to ‘account for’ or ‘explain’. One evasion, already in SS, is to insist that ‘the transformational part of grammar applies’ ‘properly’ to ‘the forms’ (‘the terminal strings’ or ‘phrase markers’) that ‘underlie’ sentences, yet to show them applying to the sentences themselves (SS 45, 47f, 66, 71, 73, 88f, 91, 107 vs. SS 35f, 42f, 74f, 78f, 80ff). In AT too, he may make ‘no careful distinction’ ‘between the basic string and sentence itself’ and proceed on the ‘simplifying and contrary to fact assumption that the underlying basic string is the sentence’, and the ‘base phrase marker is the surface structure as well as the deep structure’ (AT 18). This matter is by no means trivial if making the two levels ‘distinct’ is ‘the central idea of transformational grammar’ (AT 16f) (7.62). Surely more precision could be attained, especially if Chomsky is ‘concerned’ ‘primarily’ with ‘extremely simple sentences’ e.g., those ‘with a single element in the basis’ (AT 63, 18). He condemns ‘modern structural (taxonomic) linguistics’ for ‘assuming that deep and surface structures are actually the same’, yet says his ‘grammar does not, in itself, provide any sensible procedure for finding a deep structure of a given sentence’ (AT 16, 141) (cf. 6.38, 61; 7.7). The pathway between surface and deep is further obscured by Chomsky's habit of ‘leaving out quite a few transformations’, ‘refinements’, qualifications’, or ‘details of formalization’ he says aren't ‘relevant’ (AT 131, 92, 81f, 222).

7.81 Also, he makes use of scant data from English, and far less from other languages. Most examples are merely quoted, or are listed without further analysis as ‘grammatical’ or ‘deviant’. Only 28 invented sentences in SS and 24 in AT are analysed, i.e., broken into constituents, formalized, paraphrased or ‘transformed’. One sentence (‘“sincerity may frighten the boy”’) and its alterations gets analysed and discussed over 26 pages in the two books (SS 42f, 78, AT 63ff, 68f, 71, 73, 75f, 85f, 107, 111, 119, 149, 152f, 157, 165, 211, 228, 230) -- a lot of work for it (Humpty Dumpty would have paid it royally). ‘Diagramming’ a ‘“tree structure”‘, in which ‘a sequence of words is a constituent’ if it is ‘traced’ ‘back to a single point’ (SS 27f, AT 12), is done for one sentence in SS and 6 in AT. The trees are called ‘(base) phrase markers’ (AT 65, 69, 71, 86, 108f, 128, 130f, 171, 184), but one is called a ‘deep structure’ (AT 178) (compare the much simpler ‘deep structure’ of ‘Noun phrase -- Verb -- Noun phrase -- Sentence’ on AT 23).³⁹ Chomsky calmly remarks that ‘the interpretation of such a diagram is transparent’, and that ‘the procedure for constructing a phrase marker’ from a ‘derivation’ is a ‘minor matter of appropriate formalization and involves nothing of principle’ (AT 164, 107) (cf. 7.86).

7.82 Yet another evasion regards the role of underlying entities in the process of language understanding. Chomsky proposes to ‘judge formal theories’ by ‘their ability to explain’ and ‘clarify’ ‘facts’ about how ‘sentences are used and understood (SS 102). In SS, he stipulates that ‘understanding a sentence’ requires ‘knowing the kernel sentences from which it originates (more precisely the terminal strings underlying these)’, and raises the hope that ‘the general problem of analysing the process of “understanding”‘ might be ‘reduced’ to ‘explaining how kernel sentences are understood’ (SS 87, 92, 104n, 107f). In AT, he announces (with no mention of his earlier statement) that ‘kernel sentences’ ‘play no distinctive role’ in the ‘interpretation of sentences’ (AT 18). In return, ‘phrase markers’ are now presented as ‘the elementary content elements from which semantic interpretations of actual sentences are constructed’ -- an ‘insight as old as syntactic theory itself’ (AT 117, 221) (though he didn't know it in 1957). ‘The manner of combination provided by the surface’ ‘structure is in general almost totally irrelevant to semantic interpretation, whereas the grammatical relations expressed in deep structure are in many cases just those that determine the meaning of the sentence’ (AT 162; cf. AT 135) (cf. 7.84; 11.100; 13.47). This ‘determining’ requires that ‘the meaning of a sentence be based on the meaning of its elementary parts and manner of their combination’ (AT 162f) (but cf. 2.29; 5.67, 75f; 11.36; 12.93; 13.59). The requirement is congenial for an approach based on analysing and assembling structures, but is violated even by his own handpicked samples. His interpretation of ‘“I had a book stolen”‘ depends not on the parts and their combination, but on ‘elaborations of the sentence’ to supply contexts (like ‘“I had a book stolen from his library by a professional thief who I hired”’) -- just the recourse to ‘entirely different constructions’ he later calls ‘irrelevant’ (AT 21, 161).

7.83 A similar evasion pertains to the production of sentences. In SS, Chomsky cautions against a ‘misunderstanding’ that ‘grammars’ ‘described’ as ‘devices for generating sentences’ are ‘concerned with the process of producing utterances rather than’ ‘analysing and reconstructing’ their ‘structure’ (SS 48). The ‘synthesis and analysis’ ‘speaker and hearer must perform’ ‘are both outside the scope of the grammars’ he proposes, and (he adds without evidence) ‘are essentially the same’ -- a reason why such ‘grammars’ ‘are quite neutral as between speaker and hearer’ (but cf. 11.81, 85; 12.47). A ‘grammar is simply a description’ of the ‘set of utterances’ ‘which it generates’, and his ‘schema for grammatical description’ does ‘not permit us to state the actual processes at work’ (SS 48, AT 92). These are issues for a ‘theory’ or ‘performance model’ of ‘language use’, which ‘must incorporate a grammar’ but must not be ‘confused’ with it (AT 141, 9). Yet Chomsky himself is the one who pictures the ‘grammar’ ‘producing language’ or ‘strings’ or ‘sentences’, who refers to ‘the process of generating sentences’, and who equates ‘generating’ with ‘creating’ (SS 11f, 13, 18, 30f, 38, 45f, 103, 35, AT 135f, i.a.) (cf. 12¹³).

7.84 In AT, he again laments the ‘continuing misunderstanding’ of ‘generative grammar’ being ‘a model for a speaker or a hearer’, or for the ‘production and perception’ of ‘speech’ (AT 9). He is ‘attempting to characterize in the most neutral possible terms the knowledge of the language that provides the basis for the actual use of language by speaker-hearer’; he ‘says nothing about how’ they ‘might proceed, in some practical or efficient way’. To speak of a grammar as generating a sentence’ ‘means simply’ that it ‘assigns this structural description to the sentence’. This ‘persistent’ ‘confusion’ might ‘suggest a terminological change’, but Chomsky ‘sees no reason for a revision’, noting that ‘the term “generate” is familiar’ in this sense ‘in logic’ (e.g. ‘the theory of combinatorial systems’). But what better reason for ‘revision’ could there be than to clear up a ‘continuing misunderstanding’? Probably, the terms are retained because the conflation between theory and reality, between technical and commonsense terms, between grammar and real processes, forms a vital bridge to human language after competence gets detached from performance (which is ‘degenerate’) and deep structure from surface structure (which is ‘unrevealing’ and ‘irrelevant’) (cf. 7.25, 62